Tuesday, July 24, 2018

Longman By-Election

Cause of by-election: Incumbent resignation (ineligible under Section 44)
Outlook: Your guess is probably as good as mine.

I've finally found the time to write a detailed post about the prospects for the Longman by-election.  This won't be anywhere near as long as my Braddon guide but I think it is worth explaining why we are seeing Labor struggling in the polls, the betting and in commentary perceptions in this seat. (That said quite a few people think Labor will win Longman but lose Braddon instead.)  When I first wrote about these by-elections, I thought national polling gave Labor enough advantage to probably just hold Longman, but since then Labor's national position has declined.  Labor may still win the seat, but their position is quite fragile.

The two seats are both quite uncertain and difficult to project, but I think they are uncertain in different ways.  In Braddon there are so many unusual factors that could influence the outcome that I wouldn't be surprised greatly by anything between, say, 55-45 one way and 55-45 the other, close seat polls notwithstanding.  I think that the objective arguments slightly favour Labor holding Braddon (as on average does the recent polling, by a whisker), but they are only part of the puzzle and might well be overwhelmed by the unmeasurable factors.  In Longman, on the other hand, there seem to be fewer unpredictables.  It might not end up being super-close, but if Longman goes more than 53-47 either way I'll be surprised.

Extrapolating historic average swings and personal vote effects from the last election, even after adjusting for likely preference shifts, suggests Labor should just hold.  Polling (for what seat polls are worth) suggests a fair chance of the LNP just winning, and the seat is historically an LNP-leaning seat.  So it is very hard to pick the outcome in Longman.

General Matters

In trying to frame the Longman by-election, the first point from by-election history is that by-elections contested by Governments in Opposition-held seats have an average swing of 1.1% to the Opposition.  However, there is a relationship between that swing and national 2PP polling at the time.  Based on my present aggregate (as adjusted for One Nation preferences) the expected swing would be 1.8%.  However, voting intention has been swinging rapidly to the Coalition in recent weeks and by election day the expected benefit might be less.  (This benefit comes with a large plus and minus too, and most of the contests used to estimate it are very old.)

The second point in Susan Lamb's favour is personal vote effects.  Normally, an Opposition seat by-election would involve the loss of an Opposition personal vote.  In this case that isn't an issue, so we should expect the swing to Labor to be about a point greater, all else being equal.

There's more. Lamb has held the seat for two years and in that time would have built up some personal vote, but not as much as for a full parliamentary term.  Lamb defeated Wyatt Roy, who would have also built up some personal vote over two terms in office (although also a liability in preference terms, which I will come to later.)  The LNP candidate Trevor Ruthenberg is a former state MP for the overlapping (now abolished) seat of Kallangur, but the overlapping portion isn't very large, and his profile would have reduced over two years anyway.  So it seems fair, all else being equal, to give Lamb about 1.1 points in assumed personal vote effects (compared to the standard 1.4 points for a sophomore surge at a normal election according to Poll Bludger.)  All else being equal, we now have Lamb on for 50.8+1.8+1.0+1.1=54.7%.  (Again, plus or minus a lot.)

However, we should be very cautious about using the 2016 election as a baseline.  The reason for this is that this was one of only two cases recently where Longman has supported Labor more than the national average.  From 1996 to the present, the average result in Longman compared to the national average has been LNP +2.5, compared to ALP +1.2 in 2016.  This suggests the 2016 result was a little bit unusual, and the main reason for this was preference flow.  Longman is different in this regard to Braddon, which over the same period has nearly always leant slightly to Labor.

Another point regarding higher-level polling is that Labor in Queensland is travelling strongly, with about a 5% swing in aggregate polling.  However, as Peter Brent has observed through the years, there is a remarkably strong pattern of federal Labor polling well in Queensland specifically and then flopping on election day.  So whether that lead is even real at the moment (either statewide or specifically in a seat where there has been an intense campaign) is open to question.

One Nation And Other Preferences

Where things get difficult for Lamb is in the potential for the flow of One Nation and other minor party preferences to change.  In 2016 One Nation preferenced against incumbent Wyatt Roy, and 56.5% of their preferences went to Labor.  From a One Nation vote of 9.4%, this gave Lamb an assist worth 0.61% in swing terms, or 1.22% in terms of margin.  Had the One Nation preferences split 52-48 to him, Roy would still be in parliament.

In the Queensland election One Nation card preferences did not make a lot of difference to the outcome (which hasn't stopped an abundance of false statements blaming them for the LNP's defeat).  Flows in seats where One Nation preferenced the LNP were only about 12 points stronger than those where they preferenced Labor.  However, there was no seat where the flow to Labor was as strong as in Longman 2016.  It seems to have been one of a few 2016 special cases of One Nation deciding to give a sitting member the flick and its voters responding with relish.

To show that Longman voters are capable of passing preferences strongly from One Nation to the LNP, One Nation were eliminated in third place in 2017 in the overlapping state seats of Murrumba,  Glass House, Kurwongbah and Pumicestone.  In these, their preferences flowed to the LNP in the following proportions (in that order): 61.37%, 61.96%, 66.78%, 71.62%.   Moreover, of these, in Glass House One Nation actually preferenced Labor on their card.  I have no figures for Morayfield because there One Nation finished second.

While there is some past evidence (from 1998) that One Nation voters may not preference the Coalition as strongly at federal as at state level, there is obviously the potential at least for a 65% One Nation flow, perhaps higher.  A 65% One Nation flow off the 2016 result of 9.4% would change the 0.61% advantage to Lamb to a 1.41% disadvantage, for a 2.02% swing.

As noted, scepticism is warranted about whether the One Nation preferences will flow as strongly as suggested.  However, a further problem is that the One Nation vote in polling is much higher than in 2016 (though One Nation tends to underperform on polling day), and a string of negative reports about its candidate Matthew Stephen don't seem to be harming his vote.  The other concern for Labor in the Longman polling is that the vote for "others" parties (excluding the majors, One Nation and the Greens) is likely to be down on the 11.8% such parties polled in 2016, with One Nation picking up the slack.  The "others" split 55.5-44.5 to Labor in 2016 but are unlikely to split to Labor at all this time around, especially given the absence of the Drug Law Reform party which was a very strong ALP preference source.  So that's a further source of possible preference drift.  Even assuming the preference flow from the "others" is 50-50 this time around (and I doubt it will be), that's another 0.65% off Labor's tally, pushing Labor down to about 52-48.

On the other hand, Labor does at least have the donkey vote on its side (when oh when will we get rotating ballots for Reps elections?).  However this is probably worth less than half a percent in most electorates.

Longman Polls

The ritual disclaimers apply.  Seat polling isn't useless but is unreliable. There are many reasons why this might be so.  And if multiple public seat polls get more or less the same thing, it can still be wildly wrong, as it was in Bass, Macarthur and Lindsay in 2016 (all of these in the Coalition's favour).  But just because it's unreliable doesn't mean it will always be wrong; maybe one of these days it will be right!

We should also be cautious about a consensus of seat polls showing a very narrow lead to the LNP because at both the last two federal elections, seat polls skewed to the Coalition.  Yet in the Bennelong by-election they all overestimated Labor, some by large amounts - though this might have been caused by language difficulties in polling that electorate properly.

It is still worthwhile running through the polls that have been done in Longman briefly.  The 2PPs are those provided by the pollsters.  In the case of ReachTEL they are respondent preferences; for YouGov-Galaxy I am unsure if they are respondent or modelled (but they are certainly not last-election!). I'm indebted to William Bowe for the calculations of primaries with undecided redistributed in the case of ReachTEL.

1. ReachTEL late May 52-48 to LNP.  Primaries after redistributing undecided Labor 40.4 LNP 37.3 "independent" 5.5 Green 2.7 "other" (One Nation not a specific option) 14.1 - I think the exclusion of One Nation and inclusion of "independent" makes this poll badly flawed.

2. ReachTEL late June 50-50. (Australia Institute commissioned) Primaries after redistributing "undecided" Labor 40.4 LNP 36.1 One Nation 15.2 Green 4.5 Other 3.8

3. ReachTEL late June 51-49 to LNP.  Primaries after redistributing undecided  Labor 40.7 LNP 37.0 One Nation 15.3 Greens 3.4 Others 3.6

4. ReachTEL July 19 51-49 to LNP.  Primaries after redistributing undecided Labor 37.3 LNP 39.4 One Nation 14.5 Greens 4.4 Other 4.4

5. YouGov-Galaxy July 17-19 51-49 to LNP. Primaries Labor 37 LNP 34 One Nation 18 Greens 5 Others 6.

6. Newspoll election week 51-49 to Labor. Primaries Labor 40 LNP 36 One Nation 14 Greens 5 Others 5.  This was the first poll to poll Susan Lamb's personal rating and found her with a mildly unflattering -10 netsat (37% satisfied, 47% dissatisfied).

Assuming the Greens preferences flow as in 2016 (80.7% ALP) then the LNP are getting about the following shares of remaining preferences in these polls: 72.3%, 68.6%, 70.6%, 56.9%, 66.8%, 63.3%.  Average 66.4% - which seems a little on the high side.   Considering them individually I think a lot of the 51-49s to LNP look more like 50-50 2PPs.  I also think that the ReachTEL polls at least, possibly the Galaxy as well, have the seven "others" candidates combined too low.  In an election where the Labor, LNP and One Nation candidates are all tarnished to some degree I'd expect more than a trivial degree of voting for none of the above.

Given that the seat polls are so close to what would be expected anyway, given the issues with seat polls and given the question of whether preferences will flow quite that strongly, I'd say the average seat poll lead for the LNP is a weak signal concerning the outcome.  Even The stronger signals that are in the seat polls are that 1. even allowing for deflation the One Nation vote is likely to be up, and 2. the One Nation preferences will favour the Coalition to at least some degree.

Special Factors

So having outlined reasons why Longman could be close but still on paper just favours Labor, what else is there?

Firstly there's the circumstance of the by-election itself.  Lamb's eligibility was in doubt since at least August last year.  As with Justine Keay in Braddon there is the strong argument that her case should have been referred to the High Court earlier.  So far we have not seen any detectable blowback against MPs disqualified under Section 44, but Barnaby Joyce's re-election was not seriously contested and John Alexander resigned as soon as an issue became apparent.  My feeling is that any blowback against Lamb for hanging on so long could be muted by the well-known facts of her personal situation.

On the other side, LNP candidate Trevor Ruthenberg has faced a storm of complaint after revelations that he claimed to have a more prestigious military medal than he had actually been awarded, and it has turned out that he made this incorrect claim in multiple places.  Ruthenberg has apologised.  This is damaging, but I'm not convinced it is worth more than a few tenths of a point.  He's not Barry Urban, and a pattern of such behaviour hasn't been established (at least not in the mass media).

A potential factor in both Longman and Braddon is the "get Bill" factor.  Bill Shorten has been an entrenched Opposition Leader during a long period of what voters see as disappointing and boring politics.  He has set new records for persistent unpopularity, having now had a negative Newspoll netsat for almost three and a half years (though this is also testament to Labor's polling success over this period, without which he might well have been removed). Galaxy polls have showed voters saying they would be much more likely to support Labor if Anthony Albanese was leader (53-47 in both seats compared to 50-50 in Braddon and 49-51 in Longman under Shorten).  The by-election has no consequence for who is in government with another election less than a year away and it is a free swing for voters who want to pressure Labor to get rid of Shorten.  How many voters who might otherwise vote or preference Labor might preference the government simply to try to pressure Labor into changing its leader?  The LNP's clever "Order the Lamb, get the Bill" message is an attempt to not make the by-election about the local candidates (perhaps fearing a sympathy vote for Lamb or that she might be popular) and an attempt to make it more about Canberra.  How many voters do strategise their voting in this way?

ReachTEL polling reported on reaction to Trevor Ruthenberg's medals error issue, with 32.5% of voters in a forced-choice question (aaaarggggghhhh!) considering he did it on purpose, 26.9% calling it a careless mistake and 40.6% calling it an honest error.  This is a rather murky result since a mistake can be both honest and careless, but the middle ground is probably that it was unintentional but sloppy.  However this poll is more useful than the final Newspoll which asked voters whether the saga made them more or less likely to vote for Ruthenberg.  This form of polling is utterly worthless, primarily because partisan respondents lie, and also because voters who genuinely say they are more or less likely to vote a certain way do not say how much more or less likely.  Even if they did, when asked about an issue in isolation voters are prone to give it more prominence.  (22% said less likely and 16% said more likely - larger effect sizes have been seen in such polls where the issues were obviously trivial or even fictional.)

There are of course many campaign and policy factors that could play out unusually in Longman.  It has been difficult to get a clear sense that they will, or that the way the issues are perceived is vastly different to how it stood in 2016. The 2016 vote seems to have been about Longman voters deciding that Wyatt Roy had become the wrong fit for the seat.  It was probably also one of those periphery seats where the Turnbull Government's "innovation" message and perceived inner-city focus resounded badly, in contrast to the inner cities.

"But No Government Has Won ..."

Yes, no government has won a federal by-election from an opposition since 1920.  However, that's a misleading statistic - the more meaningful statistic is that of 27 such by-elections for which meaningful 2PPs can be obtained or estimated, the swing required to knock over Longman has been achieved in nine of them.  Actual winning has been rare because, aside from the tendency for a slight average swing to the Opposition, most of the seats weren't super-marginal in the first place.  A victory for the Coalition in Longman would not be the freak upset that the past history of governments not winning such by-elections suggests.

I may add any further thoughts I have on Longman later, and I will certainly add any new polls.  At this stage I will have live comments on all the by-elections on this site on Saturday night.

Note: This article has been extensively corrected because it was pointed out by @GhostWhoVotes that some of the swing calculations had been double-counted.


  1. Thank you once again for a very detailed analysis that is fairly easy for us political junkies to follow my friend

  2. I follow State politics fairly closely but I don't think I heard a thing about Trevor Ruthenberg in the nearly 3 years that he was member for Kallangur, not far north from where I live. A total nonentity. And dumb or dishonest (probably dumb) - the ordinary citizen may not know the difference between the Defence Medal and the Service Medal, but somebody who has received one certainly ought to! How many points do you take off a Liberal candidate for being a dumb nonentity Kevin? Not many I expect.

  3. Being obscure is not much of a hinderance for challenger candidates; I've seen a lot of mediocre performers get into federal parliament on surprisingly large swings. The record concerning Ruthenberg's personal vote as a state LNP MP is nondescript - came in on a monster swing in 2012, left (like so many others) the way he came in in 2015. The swing against him in 2015 was 1.5 points larger than the swing he got in on in 2012; if he'd been a star performer he could have done a bit better than that.

  4. Good post but soo many unknowns...as you point out..... Forgot what the margin of error is in these 800 size samples think is plus or minus 4%.all the local polls are close...how much does change in preferences of onp make? Have they grown from last vote in 2016?10% odd....if so where do they draw their votes from?most likely reduced lnp vote..if so no advantage to lnp….usually the party shares via opinion polls are very accurate... That is an alp swing of 2% plus magnified in QLD..also hard to believe that the adverse publicity has not harmed onp and lnp votes. Pick alp retain